Mythograph Atelier creates personal abstract art with AI | Keryc
Mythograph Atelier was born as a small AI studio that doesn't just generate pretty images: it tries to make abstract painting mean something to you. How does it do that? By talking, asking, and connecting shapes, colors and spaces to your ideas, emotions and references.
What is Mythograph Atelier
It's a technical and creative proposal: instead of firing off a random prompt and returning an image, the system dialogues with you. It asks brief, targeted questions, understands tastes and an intention (a quote, an emotion, a loose idea) and then builds a prompt for an image model — in tests they used FLUX — that generates an abstract painting whose visual vocabulary is tied to personal meanings.
The goal isn't just that the image is pretty. It's that when you look at that painting you can say: "this means something to me."
Three key inspirations
1) The museum and the common art experience
A visit to the IAACC Pablo Serrano and a work by Juana Francés raised the question: why does art sometimes only speak to those who already know how to read it? Mythograph Atelier starts from the idea that meaning can be accessible: an abstract composition can embody patience, change or uncertainty without text or literal illustration.
Concrete example: the author wanted a piece to convey Marcus Aurelius's line about controlling the mind. Not as text, but as an arrangement of shapes and colors that would evoke that feeling.
2) Dynamic interfaces and agents as a bridge
The second inspiration comes from the paradigm shift conversational models brought: applications can be dynamic and adapt to you in real time. Instead of static pages with fixed filters, imagine starting with a text box and the UI shaping itself to your intention.
In Mythograph Atelier the session becomes unique: the questions, visual controls and the final prompt are generated as the conversation unfolds.
3) The value of asking: the "grill me" skill
From the "grill me" skill comes the idea: don't jump straight to action, but interrogate with intent until you reach a shared understanding. Practical translation: the agent is curious but measured, asks useful questions and prioritizes answers that improve the prompt quality without tiring you.
Technical implementation (summary for developers)
Conceptual architecture:
Input: user intention (free text, quote, emotion).
Dialogue manager: conversational flow that decides follow-up questions.
Prompt constructor: converts answers into a structured prompt for the image model (prompt engineering).
Generator: image model (tests with FLUX).
Postprocess: aesthetic adjustment and metadata that link visual elements to explainable meanings.
Relevant technical aspects:
Use of agents: ChatGPT for planning and Codex as a development assistant helped speed up the prototype.
UX/AI balance: you must limit the number of questions to avoid friction; the system should be curious but efficient.
Semantic mapping: internally you can maintain a table or graph that associates shapes, textures and palettes with concepts (for example: ascending lines -> ambition; empty spaces -> uncertainty). That structure makes it easier to generate explanations you can share.
Tentative metrics: subjective satisfaction (surveys), perceived coherence between explanation and image, iteration rate (how many rounds you take before accepting a piece).
No datasets or fine-tuning details are mentioned in the initial version; development relies on prompt engineering and iterative adjustments to the conversational flow.
Results and examples
During the hackathon the author generated several early pieces. Two descriptions used as conceptual prompts were:
"A mountain is not decoration; it's patient ambition. A door gives it a second force: a threshold toward change. I want something about building meaning within confusion."
"A calm, geometric dance where lines suggest the flow of nature and the empty spaces let you choose your path."
The resulting images are still imperfect, but they began to capture the idea: compositions someone can explain to a friend by pointing out what each element represents.
Why this matters
Isn't it more interesting to have a piece you can explain and that carries a bit of your story? Mythograph Atelier opens a path for AI to make art more accessible and personal. From a technical point of view, it also raises questions about designing conversational agents that understand aesthetic intention and translate emotion into visual language.
Next challenges and improvements
Improve the quality of the final prompt to reduce randomness.
Refine the question flow to make it smoother and less intrusive.
Enrich the semantic mapping between language and visual resources.
Explore evaluation metrics that measure perceived meaning, not just visual quality.
The project is in an early stage and competing in a hackathon, but it already shows how to combine museums, conversational design theory and generative models to create abstract art with intention.
It's a simple idea with creative risk: can AI help us find those small treasures that only sometimes appear in a painting? I think it can, and the direction looks promising.
